That's mostly cross-processing in curves. Different settings for different images, but here's the key to it. In this image, I found the areas that would ordinarily be black (or nearly so) and white (or nearly so) and put sample points on them.
Notice how high the "black" RGB values are. That accounts for the hazy quality. No true blacks. And the white isn't true (neutral) white.
To get those uneven shadow values (17.66.75), pull up the bottoms of the respective color curves. So the blue channel, for instance, would be adjusted like this:
Similarly for the highlight ends of those curves. The "white" RGB values here are 254.249.230. To get those, pull down the top ends of the respective color curves. Blue is much lower than the rest. That accounts for the yellow cast in the whites.
You can check RGB values on most of these images at the brightest and darkest points and see how the curves were crossed. In this one, the red channel was left at 0 in the shadows, but G and B were both pulled way up to give a cyan cast to the blacks. When you find combinations you like, you can save them as curves presets.
Great technique peano but i get difficulties implementing this to a photo. when i sample the darkest point and the brightest point, and applying them to a photo, it turns out totally different than i'm expecting. am i missing something here?