Mathematical morphology is a very effective tool for image processing and pattern recognition applications. Some of these applications to image processing include: nonlinear filtering, noise removal, medial axis transformation, shape recognitions and smoothing, texture analysis, biomedical image processing, industrial inspection and contour detection. The worst disadvantage of mathematical morphology is the high complexity when the images are of large sizes. One approach to overcome this disadvantage is to map the morphological operations into some parallel computational arrays. Cellular Neural Network (CNN) is a technology very well fitted for implementation the mathematical morphological operations because of their parallel architecture and the local connections of cells. In the present work CNN templates is designed and tested to perform a number of morphological image processing operations with different Image size. Some of these operations use a single layer CNN while the other use mor than one layer network. The dilation and erosion use a single layer, the opening and closing use two layers network, while the skeletonization operation use eight layers CNN. These different tasks in additional to morphological filter for salt and pepper noise removal are designed and tested. The proposed CNN based morphology operators are only to be designed and implemented using VLSI technologies. For these different tasks the test of the network shown that the system are stabile at the required output in a very short time compared to that of the suggested traditional methods. The direct derivation method is suggested for the CNN template design. This method is tested for both coupling and uncoupling networks.