Actually, Web does the reverse: he says to NOT send any role that requires heavy research.
So really you can do whatever you want with your game. ALL that should be asked is to make it interesting whatever you do.
And balanced. A vote stealing mafia with the ability to be converted is far too strong (if far too fun).
A VERY basic way to balance: Think of the fastest, easiest way for the town to win..assume everything works in their favor and they get super lucky.
Now look at it and ask yourself three questions:
1. What's the chance of that happening?
2. How hard will it be for the scum to keep it from happening?
3. How many days until the game ends under that situation?
Repeat for the scum with the same questions.
If #1 ends in anything but "Almost no chance in Hades" then the side is overpowered. Even saying "well, it's a rather poor chance" is too easy. If #2 results in a "the other side really can't stop it from happening if it's actually happening." then the other side is too weak and needs a counter.
Example: Basic game.. 2 mafia, 1 cop.
Town can win in 2 nights if the cop investigates both mafia and lives. It's not THAT easy a chance but it's a real possibility. The mafia can stop it but they have to be lucky to find the cop. Overall that's too much power. Thus you add a Godfather so the cop can't just win the game. An alternative is to give the scum a rolecop so they can look for the cop while doing nightkills.
#3..well.. I believe that a traditional game should have the mafia win in 3 days if everything works out for them. That seems to balance things up nicely. As for the town, I like trying to avoid the town winning in 1 night unless they are EXTREMELY LUCKY or the scum just stunk that badly.
Btw, if you like, ExKirby, I can help with balancing once you created all of the roles.