Hi,
The Modeller objective function value does not necessarily indicate the model quality. It is more a measure of how well the model satisfies the restraints used to calculate it. The lower the value the better the satisfaction of restraints is. You can use external evaluation methods (like Procheck or ProsaII) that sometimes give a better indication of the model quality. But these methods are mostly useful for comparing models based on different templates and/or alignments. You could proceed in the following way:
1. Build models based on an initial alignment and select the one with the best (lowest) modeller objective function value.
2. Evaluate this model with external methods.
3. If errors are detected change the alignment (or template) and got to 1.
Now with respect to Procheck. Usually models build with Modeller will have good stereochemistry because it is part of the restraints used to calculate the model (CHARMM potentials). When you observe stereochemical violations in your model it usually means that something is wrong with the alignment in that region or (more rare) with the optimization. You could try to do a more thorough optimization of you model by setting MD_LEVEL = 'refine1' before calling the 'model' routine in your TOP file. If the violations don't disappear try alternative alignments for that particular region.
I hope this helps.
Best Wishes,
Roberto