Aims. We present simulations of the perfomances of the future SPHERE IFS instrument designed for imaging extrasolar planets in the near infrared (Y, J, and H bands). Methods. We used the IDL package code for adaptive optics simulation (CAOS) to prepare a series of input point spread functions (PSF). These feed an IDL tool (CSP) that we designed to simulate the datacube resulting from the SPHERE IFS. We performed simulations under different conditions to evaluate the contrast that IFS will be able to reach and to verify the impact of physical propagation within the limits of the near field of the aperture approximation (i.e. Fresnel propagation). We then performed a series of simulations containing planet images to test the capability of our instrument to correctly classify the found objects. To this purpose we developed a separated IDL tool. Results. We found that using the SPHERE IFS instrument and appropriate analysis techniques, such as multiple spectral differential imaging (MDI), spectral deconvolution (SD), and angular differential imaging (ADI), we should be able to image companion objects down to a luminosity contrast of similar to 10(-7) with respect to the central star in favorable cases. Spectral deconvolution resulted in the most effective method for reducing the speckle noise. We were then able to find most of the simulated planets (more than 90% with the Y-J-mode and more than the 95% with the Y-H-mode) for contrasts down to 3 x 10(-7) and separations between 0.3 and 1.0 arcsec. The spectral classification is accurate but seems to be more precise for late T-type spectra than for earlier spectral types. A possible degeneracy between early L-type companion objects and field objects (flat spectra) is highlighted. The spectral classification seems to work better using the Y-H-mode than with the Y-J-mode.