Template parameters of Cellular Neural Networks should be robust enough due to the random variability of VLSI tolerances and noise. Using the CNN for image processing, one of the main problems is the robustness of a given task in a real VLSI chip. It will be shown that very different tasks, as 2D or 3D deconvolution and texture segmentation, can be solved in a real VLSI CNN environment without significant loss of efficiency and accuracy when considering the low precision (about 6-8 bits) and random variability of the VLSI parameters. CNN turns out to be very robust against template noise, image noise, imperfect estimation of templates and parameter accuracy. Parameters of a template are tuned using genetic learning. These optimised parameters depend on the precision of the architecture. It was found that about 6-8 bits of precision is enough for a complicated multilayer deconvolution, and only 4 bits of precision is enough for a difficult texture segmentation at the presence of noise and parameter variances. Tolerance sensitivity of template parameters is considered for VLSI implementation. Theory and examples are demonstrated by many results using real-life microscopic images and natural textures.