Theoretically, you need to write several subroutines defining mechanical and thermal material laws. You need to incorporate necessary couplings in one  or both of thermal and mechanical fields. This is quite an advanced topic for a starter. 

Due to very large strains near the stir zone, you need to use some intelligent techniques. Typical, lagrangian formulation would probably not work even if you use frequent remeshes. An Eulerian formulation is probably better in that sense but it also has problems.

From the physical perspective, even if you are able to get rid of mathematical issues, there still is a fundamental challenge :

- We know little about the thermo-mechanical-metallurgical material behavior at very large strains (such as in the Friction stir welding). The models developed for welding processes with small deformations won`t probably be applicable in very large deformations, very fast heating rates and so on.

Thus, from the application side, I think you should not have big expectations.