I'm new in CNNs. After discovering the pre-formed deep CNNs, which can be used for feature extraction, I ask myself if there is still an application area in the field of image recognition, in which the classical classification methods (such as e.g. SVM etc.) are preferred to the CNN's?