I'm not sure I understand your question. If the fid is produced by varian software, then it will already have the necessary headers (note the plural) prepended to the data. If data is from another brand of spectrometer, I believe Varian has utilities to convert the data to their format, complete with the necessary headers.

If the fid is from another source (say a simulated or synthetic fid) then you will have to pre-pend a file header to the entire file, and a block header to each block of data. These are well described in the Varian user programming manual. If necessary, I can send you class definitions that describe the headers. These are used by SpinWorks to read Varian data but could also be used to write Varian compatible data if you have some programming experience.

I've written a binary parser for the varian fid files. Below is the output.
The global header in varian FID files define a meta data parameter S_COMPLEX.
Additionally, each FID block defines the np_cmplx, ni_cmplex,and ni2_cmplx.
I'm reading all of these values in as false !!! This implies that my fid is not complex... And I know that, in fact, my fid file is complex.

So my confusion is as follows : Why does my fid meta data appear to be telling me that my fid points are real, as opposed to complex ?

Here is the output of my program. As you can see, many of the meta data parameters look reasonable, implying that the parser basically works okay. For example, it is getting reasonable values for the number of points, the data flag, the fid flag, the number of traces, etc.... but all the boolean flags for "complex" vs "real" seem to be off.

////////////////////////////////////////
data true
fid true

There are 1 block headers in this file type. 
    np  complex false
    ni  complex false
    ni2 complex false
    nf complex false
    more false

Added 12 meta data parameters out of 12

1 : FILE_NUM_TYPE = Integer
2 : BLOCK_HEADER_SKIP_SIZE = 28
3 : FILE_NUM_RC = Real
4 : FILE_GLOBAL_FOOTER = 0
5 : FILE_DATA_ORDER = INTERLEAVED
6 : FILE_AMOUNT_OF_DATA_FILES = 1
8 : DEBUG1 = {bytesPerTrace=8192.0, bytesPerElement=4.0, elementsPerTrace=2048.0, bytesPerTrace/bytesPerElement=2048.0}
9 : FILE_BLOCK_COUNT = 256
10 : FILE_GLOBAL_HEADER = 32
11 : BLOCK_FOOTER_SKIP_SIZE = 0
12 : FILE_ENDIAN = BIG_ENDIAN

I have a varian fid which I know to be complex, but where the header parameters npcmplx,nicmplx are both valued at 0, in the first block header. Other than that the fid seems to be fine. Is there a reason for this ?