I've been something like this with Pixaki on the iPad - that allows for a reference layer which means you don't need to downsample, just fit the pixels where you want them to be accordingly. Helps in the learning of "how would this translate to pixels?" too.