Parts of the model with known structures are not strictly restrained to their PDB templates

Dear I-TASSER team,
I have submitted a job to model a human protein composed of two domains. Structures of both, N- and C- domains of human protein are known, they lack a few loops and connecting part. There is also a known structure of the complete mouse protein, which is not that much different from the human variant. I-TASSER correctly identified all three known templates: human C-terminal and N-terminal domain (had to input the pdb for that) and the mouse structure and calculated a reasonable model. However, the C-terminal kinase domain in the model does not follow known template structure precisely. Why is that and how can I make it follow the known templates? Job id is S204290.
Thank you for your answer,
Gregor