The best solution is to go back to the intermediate files and re-run the processing from there with the correct smoothing level. Filters are additive, so if you now applied a kernel of 4mm or 8mm, you'd end up with different values than you would have otherwise doing it in a single step.
One could make an argument of using
3dBlurToFWHM, but I would say go back and do it again. You could modify the afni_proc command to start at a different point. Some indication of what you'd need to do are in
this blog post. If you need more help, post your current script and we can try to modify it for you.
