I am currently awaiting Illumina methylationEPIC (850k) data and in the mean time figuring out how to assess DMRs with demo data. I am new to these analyses but I am making use of the Maksimovic et al. cross-package workflow (https://f1000research.com/articles/5-1281/v2), which I have found very useful thus far. 

I have three sample groups: healthy controls, poor treatment responders, and good treatment responders. I was wondering if it is possible to "subtract" the differential methylation seen within the controls before comparing the two treatment groups, so that I can account for a baseline/ "normal" level of variable methylation. Would this be possible in limma, minfi or any other packages? Or would it make more sense to just compare all three groups to each other?  

Thanks in advance for any assistance! 

Assuming you're fitting a linear model to the M-values, you're saying that you want to correct for the control:

corrected_poor = poor - control
corrected_good = good - control

The null hypothesis would then be:

H0: corrected_poor = corrected_good

However, some arithmetic reveals that this is the same as:

H0: poor - control = good - control
              poor = good

So you might as well just compare the two treatment groups directly, if you're interested in their differences.