Let me outline the problem. You have 80% mutant and 20% WT in your sample. Assuming the simplest possible scenario, your data now has 20% of the reads from the mutant and 80% from the WT. This should be enough of a differerence to identify differences between WT and mutant if you already have pure WT data. You should be able to see enrichment of certain transcripts.
Frankly, the assumption of the composition might not hold at all. For instance, if your RNA isolation procedure works better in the mutant vs. WT or vice versa. I have seen this happen. In my most current work, this is certainly the case. You could perhaps get a handle of how variable the number of reads in a 20:80 mutant/WT library, using qPCR. On the other hand, if you are going to put work in to making several libraries, why not sequence?
If the questions you are interested in require you to try to identify which transcripts are present in the WT but not the mutant, obviously a transcript with a high level of support in your mixed sample, might have come from WT and not your mutant.
In general, cleaning or correcting your data requires knowing what is 'suppposed to' happen to begin with. In novel situations where you are performing a genuine experiment, one has no idea what the 'true' signal is supposed to look like. So manipulations are forever suspect.
Essentially, you should be able to get away with doing a comparative analysis between WT and your other sample that's essentially enriched for mutant transcripts. I don't think there is anything you can do about the contamination without making things worse.