As a rule, it is known from a priori data that the studied field anomalies are caused by geological structures located at a certain depth below the day surface. Separation of anomalies of the observed potential field by depth and their connection with deep objects can form the basis of interpretation schemes for modeling problems. The classical method of field separation includes spectral filtering with subsequent analytical continuation of the separated anomalies. We propose an original method of height-based transformations of potential fields based on solving the inverse problem of analytical continuation of harmonic functions from a plane to the “inner” half-space. This problem is reduced to solving the Fredholm integral equation of the first kind for the Poisson integral, which can be used to represent a harmonic function in the “outer” half-space by its boundary values on the plane. The parallel algorithm for solving the integral equation is implemented on graphics accelerators using the NVidia CUDA and AMD ROCm libraries in the application software. The results of the method application are shown on the example of separation of the vertical component of the gravity field in the Bouguer reduction for the Sarginskaya area (Urals, Russia). For this territory, a detailed 3D density model was created by solving the linear inverse problem of gravimetry.