Dear Yurii,
I was 10 years in the on-wafer test business before I changed in a company where we are making calibration kits on coaxial basis.
At first forget the simulation tools. They are good, but never good enougt for this aspect and they are only so good as the input is. You have a lot of material and traces with dispersive behaviour, you get couling with the surroundings, you get temperature drift and additional dispersion with any plastic and the calibration elements are never good enought.
We developed 15 years ago a concentrated calibration kit on a PCB board to calibrate robust PCB-RF Probes in GSG and GS designs up to 4 GHz (IZI-Probe PCB, works up to 20 GHz). But the quality and bandwith was never comparable with calibration cits on ceramic. And even 15 GHz on PCB are high!
What can you do now? I believe you should use a fitting coaxial calibration cit and measure the S parameters of your complete calibration structure between well defined coaxial ports. If you have only planar ports require it additional deembeding of the connection between. Than can you use these S parameters as kit for deembedding. If you measure at different temperatures do it again under these conditions.
Good luck!
Steffen