We discuss the problem of model dependency in the phenomenology of generalised parton distributions (GPDs). We present results of our resent analyses, where we have addressed this problem. This includes the extraction of D-term, which is related to the so-called mechanical properties of the nucleon, and the analysis on the deconvolution of deeply virtual Compton scattering (DVCS) amplitudes. A new way of effectively nonparametric modelling of GPDs based on machine learning techniques will be presented as well. Getting a better grip on the control of systematic effects, our line of research will help GPD phenomenology to achieve its maturity in the precision era commenced by the new generation of experiments.