The answer to this depends on how things like shaped pulses are implemented.I only have experience of Bruker systems, so I'll make some general comments about what you need to take care of, plus some Bruker-specific stuff about implementation.

In general for (non-adiabatic) shaped pulses, the bandwidth over which they act depends on the inverse of the pulse length. Doubling the length thus halves the bandwidth, and vice versa. Of course the power level must also be changed appropriately, so if you double the length you need 1/4 of the power in watts to give the same flip angle (or 6dB less power, i.e. make the appropriate spN value 6dB higher on a Bruker system).

If the shape is used to select a single peak (e.g. in a selective 1D NOESY using a gradient spin-echo), then the required bandwidth is not field dependent, but may need to be increased slightly for wide multiplets, or decreased if you have other nearby peaks you want to miss.

If the shape is used to select a large region of the spectrum (e.g. C=O selective pulses in protein NMR), then then length has to be decreased as you increase field strength, as the required bandwidth in Hz units increases. For example going from 500 to 600, the pulse must get shorter by a factor of 5/6, because the bandwidth must increase by 6/5. This should be done for all band-selective pulses in the sequence.

On a Bruker system, the shaped pulse settings come from the PROSOL table, they are not directly in the pulse sequence. If you use the "set default pulse widths" option in EDPROSOL, the lengths of the shaped pulses that get entered will be determined by the (nominal) magnetic field strength. This will sort things out for standard Bruker sequences, and any home-made sequences that use the same conventions for pulse program writing. Note that if the the 90 degree pulse changes between samples, e.g. because of different salt concentrations, then the getprosol command can be adapted to recalculate the power level of the pulses in the table to account for this e.g. :

getprosol 1H 8.5 1.1

would recalculate all the proton pulses (decoupling pulses, shapes) on the basis that the 90 is 8.5us at a power level of 1.1dB. This doesn't change the entries in the table, just the powers coming out are recalculated. The lengths of the shapes have to stay the same, to have the right bandwidth, so the power is recalculated, to give the right flip angle.

The actual waveform of shaped pulses in general only needs to be recalculated, if there is an off-resonance frequency encoded in the shape by phase modulation. If this is the case, you can't change the pulse length, because the phase modulation would then give rise to a different off-resonance frequency. On Bruker systems, we generally create a shape to act on resonance, and for each shape there is an offset parameter SPOFFS which sets an additional frequency offset to be applied at run time, so for band-selective pulses this offset would need to be changed for different fields. For the standard Bruker bio sequences, the necessary offsets are calculated in the pulse program, but if you are writing your own sequences you need to take care of that.

For broadband adiabatic pulses, the aim is to cover the whole spectrum range equally, so one only needs to ensure that the bandwidth is still big enough. If the difference in field is small you may be OK with the same pulses, but to change the bandwidth of adiabatic pulses requires generating a new waveform. In TopSpin, there exist a set of predefined adiabatic pulses for different circumstances and different fields. For example for carbon at fields up to 600 MHz, we use a 500us CHIRP pulse with 60kHz frequency sweep for inversion, named Crp60,0.5,20.1 (the 20.1 means 20% smoothing and 1000 points defining the shape). At higher fields, we use a similar shape with 80kHz frequency sweep, Crp80,0.5,20.1.

Regarding gradient pulses, what usually matters is only the relative amplitude of all the pulses in the sequence, so in general the gradient settings are not field-dependent. One exception might be zero-quantum suppression using the Keeler-Thrippleton method which uses an adiabatic pulse in the presence of a gradient. The strength of the gradient required relates to the inversion bandwidth of the adiabatic pulse, so if you needed to change the pulse to increase bandwidth, then the optimal gradient strength would change. In addition, you may sometimes want to increase gradient powers overall, for example to improve water suppression in biomolecular sequences, but this is not directly related to field strength. On a Bruker system you would usually do this by increasing the general gradient pulse length, so you don't have to recalculate gradient ratios - just beware that you don't make any of the delays negative by doing this, as there is not much room to do so in some standard sequences (P16=1250us is usually OK).

Hope that helps!

Peter has given an excellent answer on how field/instrument/probe/solvent dependent parameters on Bruker systems are handled through the PROSOL table. I'm sure other vendors must have similar systems. I think there may be confusion on the part of some users between a pulse sequence (which is generally not field dependent) and the parameters that feed the sequence (which are dependent).

I will amplify on Peter's answer just a bit. Parameters such as offsets and referencing information are also instrument dependent, and I believe are not stored in the PROSOL table. These are instrument parameters, not probe/solvent parameters. These can, however, be stored in parameter sets. The standard parameter sets are generated for a specific instrument with the "expinstall" command. This command can convert all standard parameters sets to the current instrument based on the absolute frequency of 0 ppm for proton (e.g. 500.13 MHz on one of our systems). Note that on Bruker systems this frequency is not solvent dependent. All standard parameter sets should then have reasonable offsets and referencing information. If you are porting one of your own parameter sets between instruments at different fields, you will have to edit offset and referencing parameters. On Bruker systems, offsets and spectral widths can be entered in field independent units. e.g. setting O1P to 4.78 ppm will put you transmitter on H2O (25 deg C, neutral pH) for all fields. For 15N you could set O3P at 100 ppm.

If you are generating a new parameter set from scratch, then the easiest thing is to start with a similar standard parameter set. e.g. a COSY or NOESY parameter set for homonuclear correlation, maybe an HNCO set for 3D biochemical experiments.

Once you have read in a parameter set for your experiment, the "getprosol" command will fill in all of the correct pulse lengths etc. for the current probe/solvent combination.