You'd just need to provide a %Katakana table which resembles the %Hiragana table. This maps romaji syllables to Unicode glyph codepoints. Note that for katakana, there are a couple differences to learn, such as how to support foreign sounds like "va."