Implicit authentication schemes are a secondary authentication mechanism that provides authentication by employing unique patterns of device use that are gathered from smartphone users without requiring deliberate actions.  Contemporary implicit authentication schemes operate at the device level such that they neither discriminate between data from different applications nor make any assumption about the nature of the application that the user is currently using.  In this paper, we challenge the device-centric approach to implicit authentication on smartphones.  We argue that the conventional approach of misuse detection at the device level has inherent limitations for mobile platforms.  To this end, we analyze and empirically evaluate the device-centric nature of implicit authentication schemes to show their limitations in terms of detection accuracy, authentication overhead, and fine grained authentication control.  To mitigate these limitations and for effective and pragmatic implicit authentication on the mobile platform, we propose a novel application-centric implicit authentication approach.  We observe that for implicit authentication, an application knows best on \textit{when to authenticate} and \textit{how to authenticate}.  Therefore, we delegate the implicit authentication task to the application and let the application provider decide \textit{when} and \textit{how} to authenticate a user in order to protect the owner's personal information.  Our proposed application-centric implicit authentication approach improves accuracy and provides fine grained authentication control with low authentication overhead.  Future research in this domain will benefit from our findings to provide pragmatic implicit authentication solutions.