We present empirical data-driven machine learning algorithms for identification optical counterparts of X-ray sources to be detected in the forthcoming SRG surveys and prediction their basic physical characteristics (such as spectral class and cosmological redshift) relying on photometric data from sky surveys (SDSS, WISE, Legacy Surveys, PanSTARRS, GAIA). The proposed methods will allow one to precisely identify optical counterpart for the majority of SRG sources and measure their photometric redshifts: photo-z point predictions together with confidence intervals and full probability distribution functions P(z). Quality of photo-z predictions has been tested on spectroscopic samples of X-ray sources in the Stripe82X and XMM-XXL-N extragalactic fields.