> for classical learning algorithms

Indeed, for relatively "simple" models, SVM can get very, very close to deep learning accuracy for classification, with only a fraction of the computing time needed.