Hi, you can use loop refinement option of Modeller. You define residue rage which you have to model It will model only your N & C terminus rest part remain same. http://salilab.org/modeller/tutorial/advanced.html

I'm not very sure of this method, because modelling loops you have to select anchors at the begining and on the end of the loop. The anchors are stable, so it is not very accuracy ab initio prediction of C or N term segmrnts. Am I right?

Yes you have to select anchor residue to model loop and you can also define anchor residue for one of the end. If your terminal mising residues are not so long (5-6) then you can model it and it will be better if you do MD simulation of that model after modelling of terminal residues.

OK, 4-5 residues. I'll try.

I am not sure how modeller would be effective.. for usually the extreme termini are usually disordered and modeling disordered regions is highly unpredictable and unreliable