With the widespread application of embedded devices, embedded software vulnerabilities and oriented attacks to embedded system are also rising. As being wellknown, the hardware architecture of the embedded devices is heterogeneous, and the software stacks on them are customizable and flexible as well. Although dynamic information extraction and analysis technologies have been successfully used for general platform in vulnerability mining, malicious code detection and so on, its usage on embedded system lags behind significantly. The paper provides (1) a Cross- Platform - Dynamic Instrumentation Framework, which, by leveraging an extensible intermediate code module and a semantic constraint database, decouples the instrumentation plugins from underlying hardware and OS; (2) a set of instrumentation interfaces which may be implemented by plugins to extract runtime state of various code granularity; (3) ability of system semantic introspection to extract processes and modules information (system semantic views), by using system attribute description files and abstract state extraction functions. Compared with TEMU and DECAF, our method shows more advantages in performance of abstract state extraction, and reducing the time and memory overhead.