You can. You could, for example, adopt the view of your camera, relative to the object, select all faces with a big box select, select invert, and delete. You'll probably need to copy normals from an original copy of your mesh for your new border verts.
Note that this saves you vert count, which saves you performance on the vertex transform part of rendering, where each vertex is transformed from object space to camera space. But it doesn't really do much (anything?) for fill rate, the actual drawing of faces on the screen, because by default, backfaces will be ignored by your GPU anyways-- they'll get "culled" at an early point in the rendering pipeline.
Vertex transformation is what keeps Counterstrike under 300fps even with a very modern computer at a very low resolution. Fill rate is what makes a game get slower as you increase the resolution. They tend to operate roughly in parallel-- that is, if your performance is limited by your fill rate, because you're rendering 30 screen-sized fog planes at 4k, reducing the number of vertices you need to transform isn't going to do much to improve your framerate.
Also, be aware that it's not only the camera that determines which parts of a model you need to expose to the rendering engine. Faces can be invisible to the camera, but visible to shadowing lights, which means hidden faces can cast shadows on visible faces. There are other rendering techniques, like mirror planes, that also render from different perspectives.
If you have a fixed camera relative to your objects, both angle and field of view, and all shadowing lights are fixed relative to your objects, and your objects are well-spaced from each other, then you can go one step further, which is to just render your objects and draw them on planes. If you render a heightmap and a normal map as well as color, and a heightmap from the perspective of your shadowing light(s), you can use these to calculate 3D effects like specular lighting and shadow buffers.
Realistically, that's difficult to do with even a single shadowing light, as if your objects are locked to a single position relative to both your light and camera, they're locked to a single position relative to each other as well. But without any shadowing lights, and the use of an orthographic camera, or by accepting some error, it's doable.