afni.nimh.nih.gov

It depends on the length of your kernel. If it is short, you could use the "differential subscripts" in 3dcalc [afni.nimh.nih.gov] to implement the filtering, with a fairly complex 3dcalc command line. This would be mostly useful for a "one off" experiment.

Modifying 3dTsmooth.c would be another option.

bob cox