Hardware Trojans inserted during design or fabrication time by untrustworthy design house or foundry possesses important security concerns. These Trojans lead to un-desired change in functionality of the design and provide easy access to sensitive information. Trojans trigger attacks/malicious activities based on very rare conditions, which can evade test-time Trojan detection but can arise during long hours of field operation. In this paper we propose run-time Trojan detection architecture for a custom many-core based on Machine Learning technique. We exploit Support Vector Machine (SVM) supervised machine learning algorithms. The Data-set is generated based on many-core router behavior under normal and Trojan triggered settings. The paper targets different communication attacks triggered by Hardware Trojans, namely core address spoofing, traffic diversion, route looping attack. Support Vector Machine (SVM) algorithm has detection accuracy in the range of 94% to 97%. We implemented a framework for many-core architecture with SVM kernel while triggering Trojans based on two different conditions. To demonstrate the performance of proposed security framework, we implement a bio-medical seizure detection application as a case study. The algorithm is mapped on 64 processing cores and it takes 2.1uS to execute whereas with the proposed security framework it requires 4.8uS execution time. The Distributed Attack Detection Framework is implemented with each attack detection module having 2% area overhead.