The OP is looking at a freq range from 5 to 7 Hz, and contemplating analog tuning, so I doubt if the picosecond jitter of a GPS clock would be an issue. Give the OP a 10 MHz xtal, and follow that with a divider in the 700000 to 1000000 range (output = 14.28xxx to 10 Hz). Follow that by a div by 2 and he has a damn near perfect square wave with approximately 0.2 ppm frequency tuning resolution (accuracy depending on the xtal). You can probably do that with a dirt-cheap PIC internal counter.
Need something different? We've got NCOs, fractional dividers, noise shaping, hybrid analog/digital PLLs, etc. The bag of tricks is deep.
But given the vague requirements, I have no idea what level of sophistication or performance is actually called for, although I suspect not much.