The malware has become more advanced that the evasion techniques they have include obfuscation, polymorphism, and the mutation of the code. The present paper argues to contribute a new malware detecting system, consisting of a combination of five malware-specific machine learning tools, Malcom, a Random Forest analyzer of PE headers, a script-classifying model with black-box-based Ngam, a sequential analyzer based on GRU, and a Random Forest obfuscation detector, all controlled by a neural network control node. Our router is different to traditional ensemble outputs where a fixed weighting is applied; we cast the model outputs and operate a meta-learning architecture as features. Applied on a synthetic dataset consisting of 10,000 files (50 malicious 50 benign), the proposed method reaches 96% accuracy, 95% precision, 97% recall and an AUC-ROC of 0.98 compared to an average performance of 86% of single models and conventional ensemble methods. The framework provides a flexible and extendable basis to deal with the increasing complexity of the malware threats.