I would advocate getting a copy of 'experimental methods in RF design', it really is very good.
Have you actually tried modulating the oscillator supply? With a crystal in play the PM is typically not overly problematic and it does make for a gloriously simple QRP set (For all that you are limited to something well short of 100% negative mod).
Here is how I would skin this.
Drive the final amplifier with a square wave and modulate the supply to the final amp, then clean up with a LPF. That way the final is in class C and your power dissipation moves to the modulation amplifier. The hardcore go class E here, but that is tricky for a beginner.
You do not want to have to build a linear final amp if it can be reasonably avoided, they are never very efficient.
A crystal osc, buffer (Can be code for a gate driver chip!), and a smallish mosfet (Cdg is the figure to watch here, miller capacitance will kill you if you are not careful), with a Vds rating at least three of four times the maximum supply voltage makes a reasonable RF power source.
Output network is probably some kind of L network followed by a LPF.
Modulator can be single ended or push pull to taste.
73 Dan.