And only CS can help with this problem. If you cannot average non-linear functions trivially, you certainly can approximate it and minimize the differences. Something like a simplex method. The problem is called "Nonlinear averaging".

e.g. http://www.eebweb.arizona.edu/Faculty/chesson/Nonlinear_aver... or http://faculty.cs.tamu.edu/schaefer/research/nonlinearSub.pd... with nice graphs