The main work would be to rewrite the transliteration rules for English → * using the IPA characters as source characters. There's 107 characters + diacritics, so this will get really complex. I don't know whether Regexes work well with IPA characters.
For example, don't replace ph with πh — it would be replaced by φ once the according rule triggers.
This should be done automatically and without any hacking in the rule files required. Actually, some hacking which is already going on there might turn out to be unnecessary once this is implemented.