I am using the DrugMatrix data to do analysis. For every drug tested, there are two to three CEL data (each drug of the same dose and the same frequency is repeated in 3 rats). My approach is to get the average of these repeated CEL data before normalization. May I know is this an appropriate approach? If this is appropriate, what package or function should I use?

No, averaging isn't an appropriate approach. It's important to keep biological replicates separate during the normalization and statistical analysis, otherwise it would become impossible to assess the repeatability of the drug assays. Even if you wanted to average over replicates to get simple summaries, that would be done after normalization.

A common approach would be to process the CEL files using the affy or oligo packages and to compare drugs using the limma package.