Energy cost of crowd-sourced continuous sensing is reported to be quite high. As the number of on-board active sensors increases, complications arise due to inter-sensor interactions. The energy-cost of the Smartphones is primarily due to wireless communications (in various modes, such as, cellular radio, GPS, Wi-Fi direct, and Bluetooth) and environmental sensing using its embedded sensors in a wireless personal area network setting. The existing popular on-device-online energy-cost profilers for Android Smartphones, namely, Amobisense and PowerTutor, are energy-hungry. In this paper, we report an efficient on-demand-online profiler, called pProf, that learns from offline-precomputed model parameters to reduce the online profiling cost. We have tested our proposed technique in a customized test-bed setup comprising of the Android Smart-phones with embedded sensors that also communicate with the neighborhood sensors on smart-wearables and Sensorcon's Sensordrone platform. Our experimental measurement studies demonstrate that, compared to the popular profilers, such as Amobisense and PowerTutor, pProf consumes typically 10-15\% lesser energy. © 2016 IEEE.