In principle, it is possible without big problems. Did myself many times.

What exactly should be modeled?

MODELLER in combination with EASY MODELLER

Run sequence in PBD-Blast, if u find something significant 30-40% similarity at least, than go for homology modeling and side chain refinement. IF you get less score than opt for Ab-initio or Hybrid modelling server. You can use Bhageerath-H tool , i am attaching link below. It is a hybrid modelling tool, uses homology modelling for similar fragments and Ab-initio of dissimilar fragments. Pretty good tool.

http://www.scfbio-iitd.res.in/bhageerath/bhageerath_h.jsp

Thanks alot for the help.I want to model a protein almost 300kDA. The protein is interacting with neuronal receptors. I want to model both the mutated and normal version of the protein and then perform docking studies of it with receptors. The receptors are also big but there are experimental evidences showing the critical region.

First you should check structural similarity as i said before, you will come to know, that does your protein has similarity with any existing PDB structure. If you find good candidate, you proceed through modelling. Do check type of mutation, find out the site or AA being mutated, further investigate the role of those AA which is being mutated in normal model. you will have better idea about about both model. Problem is , you dont have any crystal structure. If normal protein would have any existing PDB structure than it could be easy to compare both structure.

Unless you find one suitable template for your protein of interest that contains all domains it will be difficult to reliably model the relative orientation of your domains to each other even if you find templates for all individual domains. Note, any errors in your model will jeopardize your docking later on. If you don't have a template very similar to your protein, I am not sure whether this is a good project as you could spend a lot of time on this without getting anything useful out of it.

Of course you can always model *something*, but if you want more than just pretty pictures you should carefully consider what you want to achieve and whether this is feasible. Sorry to be a bit more downbeat here.

Well few domains or precisely repeats of the proteins have pdb structure but not all of them.usually 1 repeat or at maximum 4 repeats have pdb structure

I agree with Ralf. You have to find a good template for your protein before you proceed to the next step. The protein weight is not a critical issue. However, there is a question: Are your protein's domains in a single chain or not? If your protein is multimeric and multidomain as well, this made another problem in your work. In my experience with MOE, ICM, Discovery Studio and Schrodinger, the last two are the best softwares for doing such kind of work.