Before removing null productions, i.e C-> epsilon.
Since C produces Ca and Cb, we add A-->a and B-->b before deleting epsilon.If we dont add,we might miss strings that include a and b. C-->CDE will become useless state,so we remove all C's. So the final grammer after adding A-->a and B-->B is
S-->aAa/bBb/EPSILON
A-->a
B-->b.
We will not delete Epsilon in S because it forms a part of language.