I can't remember where I got this, but I like the classification of mildly gifted = 97.5th percentile IQ, moderately gifted = 99th, highly gifted = 99.9th, exceptionally gifted = 99.99th, profoundly gifted = theoretically 99.999th, but practically unmeasurable. Cathy's point is good too, though - when it comes to differentiating levels of giftedness this high, more has to be taken into account than just IQ.
Those numbers sounds appealing, but given that even WISC IV normative samples are only a couple of thousand people (a quick google says 2200 in the US, less than that in several other countries), surely 99.99th is already practically unmeasurable? I'm admittedly not a statistician, but I don't see how you, for example, validate that your test can be relied upon to distinguish between two candidates one of whom is at the 99.90th and one at the 99.99th, if your norming sample had only 2 people in both those groups together!