The Gaming Den

I'm doing something similar. I started by reviewing a variety of systems, like GURPS, 3.5, 4.0, WoD, stuff I've used before, and marking down what I liked and didn't like. I then made a list of 'desired features' that I was willing to screw with things to make happen. That list has become the primary design document and created more work for me than anything else. I thought, at first (with some degree of guilt) that anything I did would just be a series of rip-offs from other systems, but now that's proven utterly impossible and I keep having to tear things down to get what I want.

Not so much classes.

Come up with a bunch of iconic characters (the flow sheet says six, which is a good minimum unless you're designing a game for a single group that isn't going to go beyond six. Fantastic! wanted to have 30, but then stalled). These aren't members of character classes, or powerbuilds, or anything. They don't even have stats. They just have things that they do, some of which are role-protected. At this stage, a "character class," if it even exists, is just a grouping of thematically-similar characters, but you don't even need to worry about grouping them into classes.

Then you just need to assign these characters into arbitrary groups of variable size (Flowsheet says groups of three and six). When you do that, make sure that even the smallest groups can easily cover all the bases, and everyone has a protected shtick in even the largest groups. See the discussion here.

Then, once you've done that, you then design your mechanics to generate these characters, provide the proper role-protection, and so on. You don't have to feel committed to character classes you came up with earlier. For example, if your Wizard archetype includes the Duke of Flames, the Star Prophet, and the Shadow Witch, and the latter two have role-protected divinations but the Duke of Flames doesn't, you may want to split off a Fire Mage class.