In this work, we leverage the features extracted from a multi-level perspective to classify functional data. We first compute pyramid polynomial features, then integrate pyramid pooling features into the framework. Additionally, a multi-level approach of randomness analysis is introduced to measure the regularity of the functional data, which is then integrated with a subsample to represent the function impact points. Finally, a pyramid view of subsequent segment distances (via Dynamic Time Warping (DTW)) is added to the system to capture the sequential relations. After mapping the functional data into tabular form, classification is done after Principal Component Analysis (PCA) and Analysis of Variance (ANOVA), via Support Vector Machines (SVM). Results are compared with state of the art (SOTA) methods. In 4 datasets, the state of the art divide-and-merge method and other powerful functional/ensemble alternatives are outperformed. Moreover, our system is not only accurate but also competitive in terms of dimensionality/complexity (feature selection is performed to align exactly with the SOTA methods).