Nonlinearity in CMOS switches has several sources. For RF
switches it's largely about the C-V swing of the S/D parasitic
diode junctions. Which basic MOS models are not great, at.
Fancier RF models can be fitted reasonably and an RFIC
company will do that.
For classic high- / medium-voltage "DC" / low frequency
analog switches, nonlinearity comes from the interplay of
common mode position with the NMOS and PMOS devices'
VT and overdrive, modulating their on-resistance. You'll see
a "bathtub" R-vs-V characteristic, not the flat line that would
be 0% distortion.
In both cases you'd like to add realistic emulation of the cause,
to the circuit. But now you're not strictly "behavioral" anymore.
If you want realism you'll be getting intimate with the transistors.
At least until you can boil the sap down to a sweet behavioral
syrup.