Intel has a perfect Application note describing 32-bit math routines.
It is called AN39 or AN40 (or AB39 or AB40). You must be able to find it on www.intel.com . It contains 32bit/16bit, 16bit*16bit and 16/32 bits add and subtract subroutines. It's all written in assembler with an HLL interface (PLM-51). You can use it to make your own routines with different var lengths.
I used it to go up to 64 bit math. (it gets a little slow though especially divide....)