0. Do you know that you actually need to do this?
I see a GPS net name. Is this only 1.5GHz? It won't mind a few 0402s. Place them close to the receiver, I suppose. Include footprints for impedance matching if possible. (Likely you'll never need to adjust them?)
1. Is this narrowband? Have you simulated the signal path?
For example, a thinner trace connecting to the offending pads does the same thing, making an LC filter of sorts. Make it resonate at Zo, Fo to eliminate return loss over whatever bandwidth the structure has.
More likely you'll simply adjust the impedance matching network slightly, and that's that.

2. This was a typo, right?:
A single 0402 pad, 0.1mm from the nearest GND plane, can be as much as 25pF capacitor
No one makes laminate with κ = 1000...

Note that the intended capacitance of 0.7mm trace length at 50 ohms is pretty damn close to 30fF, so 0402 pads really are quite a good deal.
Note that the Altera/Intel document is talking about considerable bandwidths (over 20GHz), and small gain errors (<1dB). These are relevant to extreme precision timing applications, but not very important even to most high-speed protocols. For example, USB, PCIe, SSTL, etc. all have modestly wide eye diagrams, use compensation tricks to deal with FR-4 and losses and microstrip dispersion, and are timed with separate clocks with adequate setup and hold times, or are self-clocked, immune to small changes in delay).
2. Altium doesn't have any relational mechanisms between layers, except for the very specific operations cooked into certain layers and objects. You're probably best off putting the poly cutout or keep-out in the footprint itself.
Tim